Activity Log - Demo Research

Activity Weekly

4/26/2022 - 4/29/2022

a

Do Outsourcing PPT as Spoof-Dataset PPT

b

Implement Pre-trained Models for Image Classification

c

Implement treatment for overfitting in CNN architecture

Datasets Total (items)
Low exposure 134 
Normal exposure 381
Over exposure 224
Total 739
Results b :
Model_1

Xception

 

image-1651215400050.png

image-1651213726931.png

image-1651213809478.png

Model_2

VGG-16

image-1651215907551.png

image-1651213119503.png

image-1651213129191.png

Model_3

VGG-19

image-1651215926288.png

image-1651214276556.png

image-1651214280701.png

Model_4

ResNet50

image-1651215535064.png

image-1651214581660.png

image-1651214586179.png

Model_5

ResNet101

image-1651215555948.png

image-1651214887430.png

image-1651214892215.png

Model_6

MobileNetv2

image-1651215608560.png

image-1651215121405.png

image-1651215128274.png

 

Solving_1

Regularization

 

 

Solving_2

Weight Initialization

 

 

Solving_3

Dropout Regularization

 

 

Solving_4

Weight Constraints

 

 

 

==============================================================

 

 

For Read - Bright & Sharp - Make Sample Dataset for Training

Why? :

Plan? :

Implement? :

Milestone? :

 

image-1647232455865.png

 

Referensi Paper :

  1. Sharpness and Brightness Quality Assessment of Face Images for Recognition.pdf
  2. Contrast Enhancement of Brightness-Distorted Images by Improved Adaptive Gamma Correction.pdf
  3. Face detection and the effect of contrast and brightness.pdf
  4. Haze Image Recognition Based on Brightness Optimization Feedback and Color Correction.pdf
  5. Contrast and brightness balance in image enhancement using Cuckoo Search-optimized image fusion.pdf

==============================================================

For Read - Bright- Using method Normalize, ImageDataGenerator, Adaptive Gamma Correction, Histogram Clipping

Normalize : 

image-1647586066151.png

ImageDataGenerator 1 :

image-1647586207389.png

 

ImageDataGenerator 2 :

image-1647586280034.png

 

Adaptive Gamma Correciton::

threshold = 0.5
expourse_in = 121
Data:
XRM4NWEZ - Junia Rachmah - 3276015906950001: Dimmed
XRM4WGKU - Vidia Adistia - 3674064212900001: Bright Image
XRM7DPIC - Mike noviani - 3174046904941003: Dimmed
XRNP3RAA - Ari Yunita - 3171044406840002: Bright Image
XRO4KQ7J - Lissa Nirmala - 3273194909890001: Dimmed
XROKEC4T - Tasha Kamarita - 3171085112971001: Bright Image
XRQ3CRGM - Hilda Presti Deviyanti - 3172044507820009: Bright Image
XRQGYRGB - Liesdha Nurfitrina - 3175054210880007: Dimmed
XRQLNIT6 - Yani Ratnasari - 3173074612900001: Dimmed

Input :

image-1647586381697.png

Output :

image-1647586420883.png

 

Histogram Clipping :

1. automatic detect :

alpha 1.186046511627907
beta -45.06976744186047

image-1647586556478.png

 

2. automatic detect :

alpha 1.9921875
beta -209.1796875

image-1647586696819.png

==============================================================

 

For Read - Discontinued log - Brightness with Object Detection

Training and Test Data for detection

Model Compare with 2 model input image :
ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8 SSD300: 300×300 input image, lower resolution, faster.
SSD512: 512×512 input image, higher resolution, more accurate.
image-1648190539760.png
Dataset image annotations Class :
 Lowlight, Normal, Overexpourse
Training Method  Setup Paths
install tensorflow model zoo (ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8)
install tensorflow Object Detection API
Create label map (Checkpoint)
Create Tensorflow Records (labelmap.pbtxt , test.records, train.records)
Copy Model Config to Training Folder
Update Config for Transfer Learning
Setup pipeline.config
Train the model
Evaluate the model
Load Train Model from Checkpoint
Detect from image
Pipeline Config Paramater Model SSD MobileNet V2: Value
num_classes  3
fixed_shape_resizer h:320 w:320
feature extractor ssd_mobilenet_v2_fpn_keras
box_coder faster_rcnn_box_coder
steps  10
batch_size 4

Evaluation training has not reached the desired quality

Tensorboard train evaluation
image-1648767832369.png
Result of Detection
image-1648768013174.png image-1648768041726.png
image-1648768382960.png image-1648768124968.png

Confusing matrix with tensorboard --logdir=.

Confusion Matrix
image-1649396486997.png image-1649396531033.png
image-1649396559560.png  
AP/AR IOU Area MaxDets Value
 Average Precision   0.50:0.95 all 100 0.593
0.5 all  100 0.901
0.75 all  100 0.513
0.50:0.95  small 100 -1
0.50:0.95  medium 100 -1
0.50:0.95  large 100 0.593
 Average Recall      0.50:0.95  all 1 0.6
0.50:0.95  all 10 0.683
0.50:0.95  all 100 0.683
0.50:0.95  small 100 -1
0.50:0.95  medium 100 -1
0.50:0.95  large 100 0.683

paper sheet : heatmap exploration and train to determine the color detection value of each pixel

Heatmap-based Object Detection and Tracking with a Fully Convolutional Neural Network

The visual digital turn: Using neural networks to study historical images

A Deep Residual Network with Transfer Learning for Pixel-level Road Crack Detection

A fully open-source framework for deep learning protein real-valued distances

A Decade Survey of Content Based Image Retrieval using Deep Learning

image-1649401570838.png

 

 


Revision #8
Created 29 April 2022 06:05:29 by Muhammad Hadiidtyariangga
Updated 20 August 2022 04:04:03 by Muhammad Hadiidtyariangga